With a smoother new engine taking center stage, OpenAI has once again moved the cheese for longtime users. On November 11, the company announced that it will discontinue the Advanced and Standard versions of the old voice modes in ChatGPT on December 11, 2026, betting all its development resources on GPT-Live, launched this July.

Full-duplex takes over, and the old turn-taking conversation exits

With its full-duplex architecture, GPT-Live can respond while listening to users, without waiting for them to finish; it can also hand complex tasks off to frontier text models without interrupting the conversation. By contrast, the old versions, whether Standard or Advanced, rely on a relay of speech-to-text, the model, and text-to-speech, and still require taking turns speaking. The official rationale is a strong one: since its July launch, voice usage has nearly doubled, and most voice users have already switched to the new model. The timetable has also been set—notifications will begin on November 10, the old versions will remain available until December 11, at which point they will be removed from "Settings - Voice" and users will be automatically switched to GPT-Live, while past conversations will still be preserved.

Will the GPT-4o script repeat itself?

The controversy lies in the gap in experience. Many users on X and Reddit complain that Standard mode gives more detailed answers and connects more smoothly with text context, while GPT-Live often interrupts, loses memory, and answers off-topic; some paying users threaten to cancel their subscriptions. OpenAI acknowledges that some people prefer the old style and says it is learning from that feedback to improve the new model's depth and coherence. What is most intriguing right now is whether the company will relent in the face of backlash—it previously delayed cutting off GPT-4o due to public pressure, and this time the old voice modes may well take the same backtracking path.